[BY]
Dmytro Kremeznyi
[Category]
AI
[DATE]
Jun 14, 2024
Stability AI Ltd. introduces Stable Diffusion 3 Medium, a highly advanced, open-source AI model designed for creating photorealistic images efficiently.
Stability AI Ltd. has launched Stable Diffusion 3 Medium, the latest in its Stable Diffusion 3 series. Touted as their most advanced text-to-image AI model, it's designed to produce photorealistic images with a focus on improving how hands and faces are depicted. Built on 2 billion parameters, it supports complex user prompts and styles, accurately rendering text without spelling mistakes.
The new model stands out for its ability to generate clear text and visuals from user inputs, a significant leap over previous models that often struggled with clarity. Stability AI credits these improvements to its Diffusion Transformer Architecture, which not only enhances image quality but also ensures text is spelled correctly and easy to read.
SD3 Medium is also optimized for efficiency. It’s designed to run smoothly on personal computers, including those with consumer or gaming graphics processing units. This makes advanced AI accessible without needing high-end hardware. In partnership with Nvidia, Stability AI enhanced the model's performance using Nvidia RTX GPUs and TensorRT, boosting efficiency by up to 50%.
Further collaboration with Advanced Micro Devices Inc. (AMD) has optimized SD3 Medium for AMD's devices, including accelerated processing units and consumer GPUs. This broad compatibility ensures users have diverse hardware options for running the model.
For developers, Stability AI offers SD3 Medium through its API and has made the model weights available open source. This step is aimed on encouraging global developers to experiment and innovate, fostering a collaborative environment in AI development.
Content
